Soil erosion in the Alps : causes and risk assessment

The issue of soil erosion in the Alps has long been neglected due to the low economic value of the agricultural land. However, soil stability is a key parameter which affects ecosystem services like slope stability, water budgets (drinking water reservoirs as well as flood prevention), vegetation productivity, ecosystem biodiversity and nutrient production. In alpine regions, spatial estimates on soil erosion are difficult to derive because the highly heterogeneous biogeophysical structure impedes measurement of soil erosion and the applicability of soil erosion models. However, remote sensing and geographic information system (GIS) methods allow for spatial estimation of soil erosion by direct detection of erosion features and supply of input data for soil erosion models.
Thus, the main objective of this work is to address the problem of soil erosion risk assessment in the Alps on catchment scale with remote sensing and GIS tools. Regarding soil erosion processes the focus is on soil erosion by water (here sheet erosion) and gravity (here landslides). For these two processes we address i) the monitoring and mapping of the erosion features and related causal factors ii) soil erosion risk assessment with special emphasis on iii) the validation of existing models for alpine areas. All investigations were accomplished in the Urseren Valley (Central Swiss Alps) where the valley slopes are dramatically affected by sheet erosion and landslides.
For landslides, a natural susceptibility of the catchment has been indicated by bivariate and multivariate statistical analysis. Geology, slope and stream density are the most significant static landslide causal factors. Static factors are here defined as factors that do not change their attributes during the considered time span of the study (45 years), e.g. geology, stream network.
The occurrence of landslides might be significantly increased by the combined effects of global climate and land use change. Thus, our hypothesis is that more recent changes in land use and climate affected the spatial and temporal occurrence of landslides. The increase of the landslide area of 92% within 45 years in the study site confirmed our hypothesis. In order to identify the cause for the trend in landslide occurrence time-series of landslide causal factors were analysed. The analysis revealed increasing trends in the frequency and intensity of extreme rainfall events and stocking of pasture animals. These developments presumably enhanced landslide hazard. Moreover, changes in land-cover and land use were shown to have affected landslide occurrence. For instance, abandoned areas and areas with recently emerging shrub vegetation show very low landslide densities. Detailed spatial analysis of the land use with GIS and interviews with farmers confirmed the strong influence of the land use management practises on slope stability. The definite identification and quantification of the impact of these non-stationary landslide causal factors (dynamic factors) on the landslide trend was not possible due to the simultaneous change of several factors.
The consideration of dynamic factors in statistical landslide susceptibility assessments is still unsolved. The latter may lead to erroneous model predictions, especially in times of dramatic environmental change. Thus, we evaluated the effect of dynamic landslide causal factors on the validity of landslide susceptibility maps for spatial and temporal predictions. For this purpose, a logistic regression model based on data of the year 2000 was set up. The resulting landslide susceptibility map was valid for spatial predictions. However, the model failed to predict the landslides that occurred in a subsequent event. In order to handle this weakness of statistic landslide modelling a multitemporal approach was developed. It is based on establishing logistic regression models for two points in time (here 1959 and 2000). Both models could correctly classify >70% of the independent spatial validation dataset. By subtracting the 1959 susceptibility map from the 2000 susceptibility map a deviation susceptibility map was obtained. Our interpretation was that these susceptibility deviations indicate the effect of dynamic causal factors on the landslide probability. The deviation map explained 85% of new independent landslides occurring after 2000. Thus, we believe it to be a suitable tool to add a time element to a susceptibility map pointing to areas with changing susceptibility due to recently changing environmental conditions or human interactions.
In contrast to landslides that are a direct threat to buildings and infrastructure, sheet erosion attracts less attention because it is often an unseen process. Nonetheless, sheet erosion may account for a major proportion of soil loss. Soil loss by sheet erosion is related to high spatial variability, however, in contrast to arable fields for alpine grasslands erosion damages are long lasting and visible over longer time periods. A crucial erosion triggering parameter that can be derived from satellite imagery is fractional vegetation cover (FVC). Measurements of the radiogenic isotope Cs-137, which is a common tracer for soil erosion, confirm the importance of FVC for soil erosion yield in alpine areas. Linear spectral unmixing (LSU), mixture tuned matched filtering (MTMF) and the spectral index NDVI are applied for estimating fractional abundance of vegetation and bare soil. To account for the small scale heterogeneity of the alpine landscape very high resolved multispectral QuickBird imagery is used. The performance of LSU and MTMF for estimating percent vegetation cover is good (r²=0.85, r²=0.71 respectively). A poorer performance is achieved for bare soil (r²=0.28, r²=0.39 respectively) because compared to vegetation, bare soil has a less characteristic spectral signature in the wavelength domain detected by the QuickBird sensor.
Apart from monitoring erosion controlling factors, quantification of soil erosion by applying soil erosion risk models is done. The performance of the two established models Universal Soil Loss Equation (USLE) and Pan-European Soil Erosion Risk Assessment (PESERA) for their suitability to model erosion for mountain environments is tested. Cs-137 is used to verify the resulting erosion rates from USLE and PESERA. PESERA yields no correlation to measured Cs-137 long term erosion rates and shows lower sensitivity to FVC. Thus, USLE is used to model the entire study site. The LSU-derived FVC map is used to adapt the C factor of the USLE. Compared to the low erosion rates computed with the former available low resolution dataset (1:25000) the satellite supported USLE map shows “hotspots” of soil erosion of up to 16 t ha-1 a-1. In general, Cs-137 in combination with the USLE is a very suitable method to assess soil erosion for larger areas, as both give estimates on long-term soil erosion.
Especially for inaccessible alpine areas, GIS and remote sensing proved to be powerful tools that can be used for repetitive measurements of erosion features and causal factors. In times of global change it is of crucial importance to account for temporal developments. However, the evaluation of the applied soil erosion risk models revealed that the implementation of temporal aspects, such as varying climate, land use and vegetation cover is still insufficient. Thus, the proposed validation strategies (spatial, temporal and via Cs-137) are essential. Further case studies in alpine regions are needed to test the methods elaborated for the Urseren Valley. However, the presented approaches are promising with respect to improve the monitoring and identification of soil erosion risk areas in alpine regions.